The Employment Effects of the Italian Minimum Guaranteed Income Scheme Reddito di Cittadinanza

Abstract

The Italian guaranteed minimum income programme is the topic of a heated debate scarcely supported by empirical evidence. We investigate the effects of the programme on the number of days worked by different types of vulnerable jobseekers who receive the related income transfer in 2019 and are subject to activation attempts by employment services. We use data from Tuscany, in many ways a typical Italian region. To draw causal claims, we adopt a difference-in-differences approach with multiple time periods and staggered treatment adoptions. Our findings suggest that the programme has very limited heterogeneous effects among different types of users but also that, all in all, it is rather employment-neutral, in the sense that it neither creates significant disincentives to work nor succeeds in creating sufficient employment opportunities to solve the problem of job insecurity. An anti-poverty strategy that does not feed further welfare dependency would require a qualitative leap forward in activation efforts by employment services.

Appendices

Appendix 1

This appendix is devoted to presenting the main results of the paper relying on the identifying assumption of conditional parallel trends based on a not-yet-treated group.

There are $i=1,\dots ,N$ units, $t=1,\dots ,\tau$ time periods, and $g=1,\dots ,G$ entry timings into treatment, defining entry cohorts/groups. By assumption, once a unit becomes treated, that unit will remain treated in the following periods (irreversibility of treatment). Note that g is also a time counter, and that, for each entry cohort, the $\tau$ available time periods can be partitioned into two temporal phases: every $t<g$ is before entry; every ${\text{t}}\ge g$ is after entry.

Let ${G}_{g}$ be a binary variable equal to 1 if a unit is first treated in time period g and ${D}_{it}$ be an indicator variable for whether unit i has been treated by time t (= 1) or will receive treatment in the future (= 0).

Under the assumption that there is no interference between units (Stable-Unit Treatment Value Assumption), ${Y}_{it}\left(g\right)$ is unit i’s observed potential outcome at time t, $t\ge g$, if the unit first becomes treated in time period g, while ${Y}_{it}\left(0\right)$ is unit i’s unobserved potential outcome at time t, $t\ge g$, had that unit remained untreated. Causal effects for a given entry group g are defined as the group-time average tratment effect on the treated:

$$ATT\left(g,t\right)=E\left[{Y}_{it}\left(g\right)-{Y}_{it}\left(0\right)|{G}_{g}=1\right], \quad {\text{for all}} \, t\ge g$$

(A1)

To identify the $ATT\left(g,t\right)$ we now invoke the assumption of conditional parallel trends based on a not-yet-treated group, which states that—in the absence of treatment—the post-intervention outcome variations over time of actually treated units would have been the same of that units that will take up treatment in the future and that exhibit their same observable characteristics.

For each entry group g and for each $\left(s,t\right)\in \left\{g, \dots ,\tau \right\}\times \left\{g, \dots ,\tau \right\}$ the assumption can be written as follows:

$$E\left[{Y}_{it}\left(0\right)-{Y}_{it-1}\left(0\right)|X, {G}_{g}=1\right]=E\left[{Y}_{it}\left(0\right)-{Y}_{it-1}\left(0\right)|X, {D}_{s}=0, {G}_{g}=0\right]$$

(A2)

Since potential outcomes without the treatment are observed quantities in the control group, the assumption in [A2] makes it possible to use the information from the control group to impute the missing expectation of potential outcomes.

for the treated group had it not been treated. The resulting DiD estimator is as follows:

$$ATT\left(g,t\right)=E\left[{Y}_{it}-{Y}_{ig-1}|X, {G}_{g}=1\right]-E\left[{Y}_{it}-{Y}_{ig-1}|X, {D}_{s}=0, {G}_{g}=0\right] \quad \text{for all}\, (s,t)\ge g$$

(A3)

As in the main analysis, the covariates with respect to which we will assume conditional parallel trends are: sex (female/male); citizenship (Italian/other countries); age group (up to 29 years; 30–39; 40–49; 50 years or older); the sector in which the individual predominantly gained work experience prior to the launch of the programme (agriculture, manufacturing, construction, retail/tourism, cleaners/caretakers, other services, no sector since the individual did not work); the cumulative number of days worked divided by the cumulative number of workable days and we employ the “doubly robust” estimator put forward by Callaway and Sant’Anna (2021) also in this supplementary analysis based on a not-yet-treated control group.

The aggregation of the $ATT\left(g,t\right)s$ by calendar month estimated using the approach described so far is reported in Fig. 5.

Fig. 5

ATTs by calendar month—95% CI—Monthly working days

Considering the different profiles of recipients as a whole, we estimate—aggregating the entire post-intervention period—a positive but very limited average effect of RDC on the treated equal to 0.69 monthly working days (95% Confidence Interval: 0.32; 1.05).

These results are more optimistic than those obtained under the never-treated approach, although the precision of these new estimates is not always excellent due to the dramatic reduction in the numerosity of the control group. However, comparability is limited since, with the not-yet-treated approach, we can only estimate effects for a maximum of 8 post-intervention time points, whereas with the never-treated approach we can estimate effects up to 22 post-intervention time points.

Appendix 2

See Table 3

Table 3 Recipients by prevailing sector of activity before and after the launch of the RDC programme